AskDocs MCP Server

"""Configuration loading and data structures.""" import os import sys import tomllib from dataclasses import dataclass from pathlib import Path def log(msg): """Print to stderr instead of stdout for MCP compatibility""" print(msg, file=sys.stderr) # Default models and server DEFAULT_EMBEDDING_MODEL = "snowflake-arctic-embed:latest" DEFAULT_LLM_MODEL = "qwen3:14b" DEFAULT_SERVER_URL = "http://localhost:11434" # Retrieval settings RETRIEVAL_K = 12 # Number of chunks to retrieve (higher = more context) USE_MMR = True # Use Maximum Marginal Relevance for diverse results # Chunking settings (larger chunks = more context, smaller = more precise) CHUNK_SIZE = 3000 # Characters per chunk CHUNK_OVERLAP = 600 # Overlap between chunks @dataclass class DocSource: """Configuration for a PDF documentation source""" name: str # Tool name (will be prefixed with 'search_') description: str # Tool description path: Path # Path to PDF file @dataclass class InitializedDoc: """Runtime data for an initialized documentation source""" retriever: any # LangChain retriever chain: any # RAG chain description: str # Human-readable description def load_config(config_path: Path) -> tuple[list[DocSource], dict]: """Load documentation sources from a TOML configuration file. Args: config_path: Path to the configuration file Returns: Tuple of (list of DocSource objects, config dict with settings) Raises: FileNotFoundError: If config file doesn't exist ValueError: If config file is invalid """ if not config_path.exists(): raise FileNotFoundError(f"Configuration file not found: {config_path}") log(f"Loading configuration from: {config_path}") with open(config_path, "rb") as f: config = tomllib.load(f) if "doc" not in config: raise ValueError("Configuration file must contain [[doc]] entries") doc_sources = [] config_dir = config_path.parent for doc_config in config["doc"]: # Validate required fields if "name" not in doc_config: raise ValueError("Each [[doc]] entry must have a 'name' field") if "description" not in doc_config: raise ValueError( f"[[doc]] entry '{doc_config['name']}' missing 'description' field" ) if "path" not in doc_config: raise ValueError( f"[[doc]] entry '{doc_config['name']}' missing 'path' field" ) # Resolve path relative to config file directory doc_path = Path(doc_config["path"]) if not doc_path.is_absolute(): doc_path = (config_dir / doc_path).resolve() doc_sources.append( DocSource( name=doc_config["name"], description=doc_config["description"], path=doc_path, ) ) # Extract optional model and server configuration # Priority: Environment variable > hardcoded default server_url = os.environ.get("ASKDOCS_OLLAMA_URL", DEFAULT_SERVER_URL) settings = { "embedding_model": config.get("embedding_model", DEFAULT_EMBEDDING_MODEL), "llm_model": config.get("llm_model", DEFAULT_LLM_MODEL), "server_url": server_url, } log(f"✓ Loaded {len(doc_sources)} documentation source(s) from config") log(f"✓ Using Ollama server: {settings['server_url']}") log(f"✓ Using embedding model: {settings['embedding_model']}") log(f"✓ Using LLM model: {settings['llm_model']}") return doc_sources, settings